Home · Solutions · IT & services

Solution · IT & services

The ten tasks that eat your centre's week, with hours and volumes attached

Task mining: what your shared-service desks really do

Four weeks of consented desktop capture on volunteer desks turn a backlog of automation opinions into a ranked pipeline with measured hours behind every item.

Quick winMicrosoft TeamsHuman in the loopAI where it earns its place
63automation ideas sit in this illustrative centre's backlog. None carries a measured volume, and the top of the list is whoever asked last.

Executive summary

Challenge

Your automation backlog is a wish list until somebody measures what the desks actually do all day.

What changes

Nothing is installed until the scope is agreed.

Business value

Build capacity goes to the tasks carrying the most measured hours, not to the sponsor who asks most often.

Systems involved

the ranked pipeline in UiPath Automation Hub; UiPath Insights; a Power BI read-out

Business problem

Automation discovery

Automation programmes in shared services start with a list. Workshops are run, team leads say what feels repetitive, and the answers land in a spreadsheet with a self-reported time per case and a volume nobody counted. It is honest and also wrong: people remember the tasks they dislike, not the ones that quietly consume the most hours.

Event logs close part of the gap: process mining reads what SAP or the ticket tool recorded and is good at throughput, rework and waiting. It cannot see the twenty minutes between two system entries, the spreadsheet reconciling two reports, the portal with no interface, the check added after an audit finding. That is where most repetitive work in a service centre hides.

So the pipeline is built on argument. Capacity goes to the idea with the most persistent sponsor, and estimates get defended rather than measured. Team leads lose faith after the second build that returns less than promised, and the finance director funds a roadmap whose numbers trace back to a workshop nobody minuted.

How it works today

  1. PersonA workshop per department; team leads list what feels repetitive
  2. PersonEach idea enters a spreadsheet with a self-reported handling time and a guessed frequency
  3. WaitingThe list waits for the quarterly steering meeting, where budget goes to two or three
  4. Risk of errorNo estimate is checked against reality, so the sheet becomes the business case
  5. PersonThe team builds whatever has the most senior sponsor behind it
  6. Risk of errorHalfway through the build the four ways the task is really done come to light
  7. WaitingThe rest of the backlog ages until the next reorganisation, when the workshops start again
PersonWaitingRisk of error

Why the current process costs more than it appears

The cost grows where nobody is looking.

  • Estimating from memory is wrong in both directions: a task that feels endless runs twice a week, one nobody mentioned takes forty minutes a day on every desk.
  • Variants are what break a build. A task described in one line often runs five or six ways across a team, and the versions nobody named surface during development, when redesign costs most.
  • Scarce build capacity goes to the wrong queue: a centre ranking eight candidates a year by sponsor seniority makes its most expensive decision on its least reliable data.
  • A roadmap that cannot be defended in numbers is the first line cut when budgets tighten, however good the ideas on it were.

Cost of inaction

Twelve months of a pipeline ranked by who asked loudest≈ €294,000
Three planning cycles decided the same way≈ €882,000
The same ratio across all 240 desks in the centre (per year)≈ €1,764,000

The ranking decides which seven ideas in eight never get built, and the ranking has no numbers in it. The centre keeps delivering a few automations a year and reporting savings nobody can reconcile with a payroll line, because the baseline was an estimate to begin with. The backlog grows faster than the build capacity, so the ranking matters more every year while the basis for it stays the same.

The quieter cost is credibility. A programme that twice promised more than it returned has its next business case read sceptically, and the tasks with the largest measured hours, usually not the loudest, stay in the queue behind them.

Illustrative scenario

A plausible organisation with realistic proportions. The figures are there to be recalculated on your data; they are not a client result.

Organisation

A shared-service centre of 240 people in Poland serving four European entities of an insurance group: accounts payable, accounts receivable, master data and HR administration, on SAP S/4HANA, two broker portals and Microsoft 365 E3. A three-person automation team already runs UiPath in an EU-region tenant.

Volume

A four-week capture on 40 volunteer desks, about one in six, giving roughly 4,800 hours of recorded screen time; the analysis returns twelve recurring tasks with 34 variants.

Current process

Ideas arrive through a Teams channel and one workshop round a year. Sixty-three are open, each with a title, a sponsor and a self-typed estimate.

Bottleneck

The team builds six to eight automations a year, so seven ideas in eight never happen. Nothing carries a counted number, so the ranking is a negotiation, and two of last year's builds returned less than promised.

Solution

A consented capture with an agreed application list and automatic masking; task graphs and variants from UiPath Task Mining; candidates in UiPath Automation Hub with effort and frequency attached; the roadmap reviewed monthly in Teams.

Potential outcome

The twelve tasks account for 38% of captured desk time in the model, of which 45% is realistically automatable: roughly 11,300 hours a year across the 40 desks. Illustrative figures, not a client result.

Proposed solution

Nothing is installed until the scope is agreed. We write what will and will not be recorded, prepare the note that goes to the teams, and take the scope through the works council, trade union or employee representative where one exists. Only applications on the agreed list are captured, so private mail, banking and HR self-service stay outside the recording.

UiPath Task Mining then records the volunteer desks, masks personally identifiable information automatically during processing, and clusters the traces into task graphs and variants, so a task described as one thing in a workshop appears as the five ways it is really done, each with its own frequency and time. Candidates leave as a Process Definition Document and a Studio skeleton for UiPath Automation Hub, scored there on effort, benefit and readiness.

Where event logs exist we set the finding beside UiPath Process Mining: task mining shows how work is done on screen, process mining how a case moves between systems. This replaces nothing. Our KYP.ai process X-ray measures a whole centre and returns a process map; this capture goes deeper on a short list of desks, inside the UiPath tenant you already have.

Native capabilities used

UiPath Task Mining recorder (visible application list, user pause control, automatic PII masking in processing, task graphs and variants, Process Definition Document and Studio skeleton output, submission to Automation Hub); Automation Hub idea pipeline and cost-benefit scoring; UiPath Process Mining; UiPath Insights dashboards and CSV export; Microsoft Teams channel and Approvals app

What we build

The consent and communication pack, the capture design and application list, the task taxonomy in business language, the scoring model agreed with finance, the Automation Hub configuration and the Power BI read-out

Custom integration

None for the capture; the read-out is assembled in Power BI from Automation Hub and Insights exports

How the automated process works

  1. PersonScope, consent and the application list are agreed with team leads and employee representatives before anything is installed
  2. SystemThe recorder reaches the 40 volunteer desks through your standard software channel; each person sees what is captured and can pause it
  3. AutomationActions and application context are recorded and uploaded; personal data is masked automatically during processing
  4. AutomationTask Mining clusters the traces into task graphs and variants, with the frequency and time each variant takes
  5. PersonAutomation team and team leads name the tasks in business terms and drop what is out of scope
  6. AutomationSelected tasks enter Automation Hub as candidates with their documentation, scored on effort, benefit and readiness
  7. PersonThe roadmap is agreed in the Teams channel and approved once a quarter by the steering group
PersonSystemAutomation

Human-in-the-loop model

Automation handles

  • Recording the agreed applications on the agreed desks, and masking personal data in processing
  • Clustering thousands of actions into task graphs, variants, frequencies and effort
  • Producing the documentation and the Studio skeleton for each candidate and carrying it into Automation Hub

People decide

  • Whether to take part at all, and when to pause the recorder on their own desk
  • Which applications are in scope, agreed before capture with leads and employee representatives
  • Which tasks are real candidates and how the roadmap is ranked, owned by the steering group

Before and after

BeforeAfter
Basis for the roadmapself-reported workshop estimatesmeasured hours and frequency per task variant
Backlog items with a counted volumenone of the 63every candidate reaching the steering group
What the developer receivesa one-line idea and a recollectiondocumentation, a variant map, a Studio skeleton
When variants surfaceduring the build, at redesign costbefore the estimate is signed

Systems and integrations

Where a rule suffices we do not use a model. Where judgement is needed, a person decides.

Inputs

  • 40 consented desks in accounts payable, accounts receivable, master data and HR administration
  • the agreed application list
  • SAP event logs where they exist

Automation layer

  • UiPath Task Mining
  • UiPath Automation Hub
  • UiPath Process Mining
  • UiPath Automation Cloud (EU region)

Target systems

  • the ranked pipeline in UiPath Automation Hub
  • UiPath Insights
  • a Power BI read-out

Human touchpoints: task review sessions with team leads; the roadmap channel in Microsoft Teams; the quarterly approval in Teams Approvals

40 consented desks in accounts payableUiPath Task MiningUiPath Automation Hubthe ranked pipeline in UiPath Automation Hubtask review sessions with team leads

Technologies used

UiPath Task Mining

records the captured desks, masks personal data in processing, clusters actions into task graphs and variants with time and frequency

A
UiPath Automation Hub

receives candidates with their documentation, scores effort against benefit, holds the ranked pipeline

A
UiPath Process Mining

where event logs exist, sets the desk-level finding against system throughput and rework

A
UiPath Insights

shows what delivered automations return, so the next scoring round uses measured benefit

A
Microsoft Teams (channel and Approvals app)

candidate intake, monthly roadmap review, quarterly sign-off

A
Power BI

the steering-group read-out: hours by task, automatable share, pipeline by quarter

A
Averified product capability (vendor documentation)

Illustrative economic model

Start by questioning the assumptions.

Illustrative model
40 recorded desks × 1,650 productive hours a year= 66,000 h a year
66,000 h × 38% in the twelve recurring tasks≈ 25,100 h a year
25,100 h × 45% realistically automatable≈ 11,300 h a year
11,300 h × €26 fully loaded hourly cost≈ €294,000 a year
Annual illustrative value pool across the captured desks≈ €294,000

None of these figures is a client measurement; they are this illustrative centre's assumptions, and the capture replaces them with counted values. We assume 40 captured desks at 1,650 productive hours a year, 38% of recorded desk time attributed to the twelve recurring tasks, and 45% of those hours realistically automatable once judgement steps, rare variants and systems due for replacement are removed. €26 is a fully loaded hourly cost for a shared-service role in Central Europe. The result is a value pool to prioritise against, not a saving.

Business benefits

  • Build capacity goes to the tasks carrying the most measured hours, not to the sponsor who asks most often
  • Estimates arrive with their variants attached, which removes the most common cause of overrunning builds
  • Developers start from documentation and a Studio skeleton generated from real traces
  • Small tasks leave the list early, so nobody spends a quarter automating a few hours a month
  • Teams see their own work in numbers and have a say in what is automated first

The management view

  • Automation spend becomes traceable: every roadmap item carries the hours it was measured at
  • The programme can answer the board's question about return before the first robot is built
  • Capacity planning improves, because the centre knows what share of desk time is repetitive work per team
  • The pipeline outlives the people who built it, sitting in Automation Hub with its evidence rather than in a deck

Board-level KPIs

share of backlog items with a measured volumeautomatable hours identified per capturehours released per quarter after go-livetime from candidate to productionvariance between estimated and delivered benefit

Security and governance

Where the data sits and who can see it.

  • Participation is voluntary and reversible. Everyone on a captured desk sees the full list of applications being recorded and can pause the recorder at any time, within the documented limit of one hour per pause.
  • Scope is narrowed before capture rather than filtered afterwards: administrators and users define which applications are captured, so private mail, banking and HR self-service never enter the recording.
  • Personal data is masked automatically during processing, findings are reported at task level, and we commit contractually to produce no per-person output.
  • Captured data stays in your UiPath Automation Cloud tenant in the EU region, encrypted in transit and at rest, with access and deletion controlled from the administration console.
  • Where a works council, trade union or employee representative exists, purpose, scope and retention are agreed with them before installation. The note to the teams goes out before the recorder does.

Why now

01

Build capacity for next year is being allocated now, from the list that exists. On the modelled numbers the pool behind 40 desks alone is around €294,000 a year.

02

Shared-service scopes keep growing faster than headcount budgets, and the standard response, another round of workshops, produces the same unmeasured list as last time.

03

Task Mining now runs from UiPath Automation Cloud only, after the Automation Suite service was retired in October 2025, so nothing on-premise has to be stood up.

Relevant executive roles

Head of Shared Services

The roadmap becomes a ranked list of measured hours per team instead of a negotiation

CFO

Business cases carry a measurable basis before money is committed, and delivered benefit can be compared with what was promised

CIO

The automation team spends its capacity on the highest-return tasks and builds from documented variants

CHRO

Work is measured with consent, an agreed scope and no per-person reporting

Common questions and objections

Isn't this surveillance of our people?

It measures tasks, not people, and it is built to be refusable. Participation is voluntary, everyone sees the applications being recorded and can pause the recorder, capture is limited to applications agreed in advance, and personal data is masked automatically during processing. We report at task level and produce no per-person output.

We already ran the KYP.ai process X-ray. Is this not the same thing?

They answer different questions. The X-ray measures where work sits across the centre and returns a process map; this capture goes a level down on a short list of desks and returns click-level variants, effort per variant and a document a developer can build from.

Will people work differently because they know they are recorded?

Some will, in the first days. That is why the capture runs four weeks rather than four days, why the first week counts as settling in, and why we compare task frequencies with system counts wherever event logs exist.

When this is not the right solution

  • Work that is genuinely different every time, such as complex case handling or design. The capture will find variants and very little repetition, and the report will say so.
  • Sites where the employee-representation route would take longer than the value of the finding. There we start from system event logs and come back to desktop capture later.
  • Centres with no capacity to build anything in the next twelve months. A measured pipeline nobody can act on documents the problem at a cost.

A question for the next management meeting

Of the automation ideas we intend to fund next year, how many carry a counted number of hours, and who counted them?

Implementation approach

We start with one slice of the process and extend only once it is proven.

We deliver

  • The consent and communication pack: what is recorded, what is not, who sees the output, how long data is kept
  • The capture design: which teams, which desks, which applications are on the list
  • Recorder deployment with your IT team, plus a short dry run before the real capture
  • Four weeks of capture with weekly coverage checks, so a silent desk is noticed within days
  • The analysis: task graphs, variants, effort and frequency, named with the team leads
  • The pipeline in Automation Hub and the Power BI read-out for the steering group

We need from you

  • Team leads for two half-days of naming and scoping the tasks the analysis returns
  • An IT route to install the recorder, and the application inventory behind the list
  • The employee-representation contact per site, early enough to agree before installation
  • Whoever owns the automation budget, present at the read-out

Stages

Consent and scope

Capture plan, application list, employee note, works-council conversation

Setup

Tenant configuration, recorder deployment, a short dry run

Capture

Four weeks of recording with weekly checks on coverage and quality

Analysis

Task graphs, variants, effort and frequency, named with the team leads

Pipeline

Candidates into Automation Hub with scoring; roadmap agreed in Teams

Quick win. Effort follows the number of teams and applications in scope, the employee-representation route in each country, and how much work already leaves an event log.